AI LMS · Learning with an AI lecturer

Build a course in minutes, taught by voice by an AI lecturer

Paste a document, a process or a job description. The AI LMS writes the slides, teaches each part by voice, answers questions, quizzes and grades, then shows you who studied, how far they got and what they scored.

Learners only need a link · Nothing to install

Features

A complete learning system, with AI doing the heavy lifting

Writing, teaching, testing and tracking learners, all in one place.

From material to course

Paste text, upload PowerPoint or PDF, or just write a request. The AI turns it into detailed material, then into slides with diagrams, pictures and notes.

An AI lecturer who teaches by voice

It teaches slide by slide, stops when a learner asks, answers, then carries on. Learners ask by voice or by typing.

Quizzes and grading

Questions from the lesson; the AI grades each answer and notes strengths and what to review. Set a pass mark, allow a retake.

Know who is learning

Learners sign in or type their name. See who took which lesson, for how long, how far and with what score, and download it as CSV.

Integration

Connect Open edX, Moodle, Canvas

Already run a learning system? Bring AI lessons into your courses with the LTI 1.3 standard.

Set up the connection →
  • Learners open the lesson from their course, with no new account
  • Each session carries the learner's name, shown in your list of learners
  • Quiz scores go to the course gradebook
  • Set it up once for the whole system, then add lessons one by one

AI Voice · Vietnamese speech AI platform

Speech AI built for Vietnamese, ready for business

Accurate, real-time Vietnamese speech-to-text, text-to-speech and LLM — from a team that has deployed voice AI for hundreds of enterprises in Vietnam, the USA, Mexico, the Philippines and Thailand.

10.53%average word error rate on 4 Vietnamese test sets
690,517 hof Vietnamese speech data collected
24 languagesfor translation and multilingual meetings
Freeevery service, no card

Internal figures, measured on data not used for training — see how they are measured

Three models, one platform

One account, one API key for every service.

Speech-to-TextVietnamese speech, transcribedReal-time streaming over WebSocket, files over REST
Text-to-SpeechAI voices that sound humanVietnamese and English voices, streamed from the first word
aivision-L1.0A Vietnamese large language modelOpenAI-compatible API — change the base URL and go

Deployment experience in Vietnam USA Mexico Philippines Thailand

Accuracy

Accuracy you can measure

Word error rate (WER, lower is better) of AIVISION's model on four Vietnamese test sets, including medical conversations and real business meetings.

VLSP
8.3%WER
5.1%CER

VLSP test set (1,000 utterances, 1.76 h, human references).

See the WER on four test sets
Test setContentAIVISION WER
FLEURS-viRead speech 3.60%
VIVOSRead speech, many speakers 4.71%
ViMedCSSMedical, with English terms 14.10%
AIV meetingsReal business meetings 19.71%
Average10.53%

AIVISION internal results on data not used for training. ViMedCSS: medical conversations with English terms; AIV meetings: real business meetings.

Why AIVISION

Built for Vietnamese

A 690,517-hour Vietnamese speech corpus, trained on a 24× H200 GPU cluster for over 500 hours — tones, three regional accents, fast speech, English mixed in.

Real time

Words appear while the speaker is still talking; AI voices start on the first word.

International experience

Deployed for hundreds of enterprises in Vietnam, the USA, Mexico, the Philippines and Thailand.

Industries

Solutions by industry

  • Education & e-learning
  • Contact centers
  • Healthcare
  • Foreign companies in Vietnam
See every industry
  • Banking & finance
  • Media & journalism
  • Public sector
  • Retail & e-commerce

For developers

Integrate in minutes

Get an API key in the console, send audio, get text back. The language model speaks the OpenAI API — switch the base URL and go.

API docs →
# Speech-to-Text: a file in, text out
curl https://s2speech.com/v1/transcribe \
  -H "Authorization: Bearer $AIVISION_KEY" \
  -F file=@cuoc-hop.mp3 -F language=vi

# LLM: OpenAI-compatible
from openai import OpenAI
client = OpenAI(base_url="https://s2speech.com/v1", api_key=AIVISION_KEY)
reply = client.chat.completions.create(model="aivision-L1.0",
    messages=[{"role": "user", "content": "Tóm tắt cuộc họp này..."}])
Dr. Giang Vo

The person behind AIVISION

Dr. Giang Vo

CEO, AIVISION

Dr. Giang Vo leads AIVISION in building a speech AI platform for Vietnamese — from the data and recognition models to the voices and the apps — and works directly with customers on their deployments.

📞 0981 419 967 · ✉️ giang.vo@aivgroups.com

Blog

Latest articles

Blog →

AI LMS

Start building your AI course

Make your first lesson in minutes and send learners the link, or talk to Dr. Giang Vo directly.

Completely free
  • Every service and app is free
  • No card, no top-up
  • The daily limit resets at midnight; need more, contact an admin

Frequently asked questions

Does it cost anything?

No. Every service is free, no card needed. Each account has a daily usage limit that resets at midnight; if you need more, contact an admin.

What is AIVISION s2speech?

s2speech is AIVISION's speech AI platform: Speech-to-Text, Text-to-Speech and the aivision-L1.0 language model, plus ready-made apps such as Voice Note, Live Translate and Multilingual Meetings.

How accurate is AIVISION Speech-to-Text?

On four Vietnamese test sets (FLEURS-vi, VIVOS, ViMedCSS and real business meetings) AIVISION's model averages a 10.53% word error rate, measured on data not used for training.

Which languages are supported?

Vietnamese first, plus English; the translation and meeting apps support 24 languages including Spanish, Filipino, Thai, Chinese, Japanese, Korean and Indonesian.

How do I integrate it?

Create an API key in the console and call the REST or WebSocket API. The language model is OpenAI-compatible, so familiar SDKs work as they are.